This chapter of R for Health Data Science by Dr J H Klopper is licensed under Attribution-NonCommercial-NoDerivatives 4.0 International
Libraries
Introduction
Data visualization is the process of representing data or information in a graphical or pictorial form. It is an important tool for understanding and communicating data and helps to make sense of large and complex data sets. Data visualization allows us to see patterns, trends, and relationships in data that might not be immediately apparent when looking at raw data. It also helps to identify trends, outliers, and other important features in the data.
There are many different types of data visualizations, including bar charts, line graphs, scatter plots, and pie charts. The choice of visualization depends on the type of data and the story you want to tell with the data.
Data visualization is an important part of data analysis and is used in a variety of fields, including public health and the biomedical sciences.
The R language has built-in functions for data visualization. We will encounter some of the in this notebook, but the emphasis is on the much more powerful ggplot2 library.
The ggplot2 library
The ggplot2 library is a data visualization package for the R programming language. It was created by Hadley Wickham and is an implementation of Leland Wilkinson’s Grammar of Graphics—a general scheme for data visualization which breaks up graphs into semantic components such as scales and layers. The ggplot2 library is designed to be easy to use, modular, and extensible. It provides a wide range of statistical and graphical techniques, including linear and nonlinear modeling, classical statistical tests, time-series analysis, classification, and clustering.
With ggplot2, you can create a variety of plots including scatterplots, line plots, bar plots, and box plots. You can also customize the appearance of your plots by modifying the themes and aesthetics. ggplot2 is a popular choice for data visualization in R because it is easy to learn and produces publication-quality plots.
Data
To illustrate the use of the built-in and the ggplot2 plots, we import a data set using the read_csv function from the readr library. The file, PlottingData.csv, is in the same folder as this Quarto document.
We use the head function to display the first six observations.
# A tibble: 6 × 9
Ventilation Age dBP CRP HB HR Diabetes Obesity Grade
<chr> <dbl> <dbl> <dbl> <dbl> <dbl> <chr> <chr> <chr>
1 Yes 69 75 105. 8.6 120 None No I
2 Yes 73 87 119. 11.2 90 None No I
3 Yes 53 91 99.9 12.4 164 None No I
4 Yes 74 100 101. 7.6 101 Type II Yes I
5 Yes 69 83 135 12.2 102 Type I Yes II
6 Yes 54 88 133. 12.9 113 Type II No II The data contains information on the age, diastolic blood pressure, c-reactive protein levels, hemoglobin levels, heart rate, diabetes, obesity levels, and grade of comorbid disease for 312 observations. The binary target variable, \texttt{Ventilation}, indicates if a person was mechanically ventilated in the intensive care unit.
Visualizing continuous data types
When we consider a single continuous numerical data type variable, the best visualization is by using histograms, box-and-whisker or violin plots, and .
When we want to visualize pairs or more numerical variables, the best visualizations are scatter and bubble charts.
Histogram
A histogram is a graphical representation of the distribution of a data set. It is a graph that displays the frequency or number of observations within specific ranges or bins. The horizontal axis of a histogram represents the range of values in the data, and the vertical axis represents the frequency or count of observations within each range.
Histograms are useful for understanding the shape and spread of a data set and for identifying patterns and trends in the data. They are particularly useful for visualizing continuous data, such as measurements or observations, rather than discrete data, such as counts or categories.
To create a histogram, you first need to choose the range of values or bins to group the data into. You then count the number of observations within each bin and plot the count on the vertical axis. The resulting graph will show you the distribution of the data, with tall bars representing a high frequency of observations and short bars representing a low frequency. You can also use histograms to compare multiple data sets by plotting them on the same graph.
The built-in hist function generates a default histogram. We stipulate the data frame object and column name attribute in Figure 1.
We notice many problems with this plot. First and foremost is the title, which is taken from the data frame object and column attribute. We use the main argument to change this, shown in Figure 2.
Next, we use the xlab argument to create a more appropriate label for the horizontal axis. We also use the ylab argument to overwrite the current vertical axis label, shown in Figure 3.
The axes ticks can also use some improvement. We use the las argument, set to 1, to print the vertical axis tick marks upright. We also use the xlim and ylim arguments to set better intervals in Figure 4.
hist(
dfr$Age,
main = "Distribution of age values",
xlab = "Age [years]",
ylab = "Count",
las = 1,
xlim = c(20, 90),
ylim = c(0, 70)
)
We can also choose the interval width by using the breaks argument. Below, we use the seq function to generate a sequence of breaks. We have to increase the limits of the ylim argument to accommodate the higher frequency of values in each bin in Figure 5.
hist(
dfr$Age,
main = "Distribution of age values",
xlab = "Age [years]",
ylab = "Count",
las = 1,
xlim = c(20, 90),
ylim = c(0, 120),
breaks = seq(from = 20, to = 90, by = 10)
)
The ggplot2 library builds a plot step-by-step using tidyverse principles. We will use the magrittr pipe throughout. We start by using the ggplot function. The first argument is the aesthetics, aes. This stipulates the variables to be used and which below to which axis. To this, we add the type of plot. Note how we build the structure of the plot by adding elements with the + symbol. The geom_hist function takes the data and creates a histogram with it. We use the breaks argument and build a sequence of breaks. Alternatively, we can use the binwidth or bins arguments. The former states the width of each bin and the latter the number of bins. Depending on the range of values in the variable, these almost never align as well as when we specify the breaks. Finally, we add the closed argument. It takes values "left" and "right". We use it to specify if the bin intervals are left or right closed. The plot is generated in Figure 6
dfr %>% ggplot2::ggplot(
aes(
x = Age
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left"
)
We add a title and subtitle using the labs function in Figure 7.
dfr %>% ggplot2::ggplot(
aes(
x = Age
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left"
) + labs(
title = "Age distribution",
subtitle = "Includes the ages of all observations"
)
In Figure 8 we add labels to the axes as well.
dfr %>% ggplot2::ggplot(
aes(
x = Age
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left"
) + labs(
title = "Age distribution",
subtitle = "Includes the ages of all observations"
) +
xlab("Age [years]") +
ylab("Frequency")
Themes allow us to create an overall look to a plot. The ggplot2 library themes can be extended using the ggthemes library. In Figure 9 we use the theme_fivethirtyeight theme.
dfr %>% ggplot2::ggplot(
aes(
x = Age
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left"
) + labs(
title = "Age distribution",
subtitle = "Includes the ages of all observations"
) +
xlab("Age [years]") +
ylab("Frequency") +
ggthemes::theme_fivethirtyeight()
In Figure 10 we specify the color of the outlines of each bar in the histogram, as well as the fill color.
dfr %>% ggplot2::ggplot(
aes(
x = Age
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left",
color = "black",
fill = "gray"
) + labs(
title = "Age distribution",
subtitle = "Includes the ages of all observations"
) +
xlab("Age [years]") +
ylab("Frequency") +
ggthemes::theme_fivethirtyeight()
We can even add some statistics to a plot. In Figure 11 we add a red, dashed vertical line to indicate the mean age.
dfr %>% ggplot2::ggplot(
aes(
x = Age
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left",
color = "black",
fill = "gray"
) + labs(
title = "Age distribution",
subtitle = "Includes the ages of all observations together with the mean age"
) +
xlab("Age [years]") +
ylab("Frequency") +
ggthemes::theme_fivethirtyeight() +
geom_vline(
xintercept = mean(dfr$Age),
color = "red",
linetype = "dashed",
size = 1
)
dfr %>% ggplot2::ggplot(
aes(
x = HB
)
) + geom_histogram(
breaks = seq(from = 0, to = 20, by = 1),
closed = "left",
color = "black",
fill = "gray"
) + labs(
title = "Hemoglobin distribution",
subtitle = "Includes all observations together with the mean hemoglobin level"
) +
xlab("Hemoglobin [years]") +
ylab("Frequency") +
ggthemes::theme_fivethirtyeight() +
geom_vline(
xintercept = mean(dfr$HB),
color = "red",
linetype = "dashed",
size = 1
)
The ggstatsplots and other similar libraries are very good at producing publication-ready plots. In Figure 12 we see such a plot as an example of using the gghistostats function. The bf.message keyword argument is set to FALSE to suppress output of Bayesian statistics. This leaves only output for frequentist statistics. The test.value keyword argument is set to the value 55, to indicate the mean under the null hypothesis.
dfr %>% ggstatsplot::gghistostats(
x = Age,
bf.message = F,
test.value = 55
) +
labs(title = "Age distribution of all the participants")
There is more information about the gghistostats function here.
The ggplot2 library allows us to generate more than one histogram, grouping a numerical variable by the levels of a categorical variable. In Figure 13 we see the distribution of ages, grouped by the two \texttt{Obesity} levels. The position = "idenity" argument and value plots the two histograms on top of each other.
dfr %>% ggplot2::ggplot(
aes(
x = Age,
color = Obesity
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left",
fill = "gray",
alpha = 0.2,
position = "identity"
) + labs(
title = "Age distribution for each obesity level",
subtitle = "Includes the ages of all observations"
) +
xlab("Age [years]") +
ylab("Frequency") +
ggthemes::theme_fivethirtyeight()
In Figure 14, we stack the histograms using the position = "stack" argument and value.
dfr %>% ggplot2::ggplot(
aes(
x = Age,
color = Obesity
)
) + geom_histogram(
breaks = seq(from = 20, to = 90, by = 5),
closed = "left",
fill = "gray",
alpha = 0.2,
position = "stack"
) + labs(
title = "Age distribution for each obesity level",
subtitle = "Stacked histogram"
) +
xlab("Age [years]") +
ylab("Frequency") +
ggthemes::theme_fivethirtyeight()
The ggstatsplot library can also produce a comparative histogram using the grouped_gghistostats function. In Figure 15 we create a histogram of the HR variable grouped by the classes of the Ventilation variable.
dfr %>% ggstatsplot::grouped_gghistostats(
x = HR,
bf.message = F,
grouping.var = Ventilation,
test.value = 100,
plotgrid.args = list(ncol = 2),
annotation.args = list(
title = "Distribution of heart rate values",
caption = "Data grouped by ventilation status"
)
)
Box-and-whisker plot
A box-and-whisker plot, also known as a box plot, is a graphical summary of a set of quantitative data. It is used to display the distribution of the data, including the median, quartiles, and range. The box plot is a useful tool for comparing different sets of data, identifying outliers, and understanding the spread of the data.
A box plot consists of a box, which represents the interquartile range (IQR), and whiskers, which extend from the box to show the range of the data. The median of the data is represented by a line drawn within the box. The lower and upper quartiles are represented by the lower and upper edges of the box, respectively. The minimum and maximum values of the data are represented by the lower and upper ends of the whiskers, respectively. Outliers, which are data points that are significantly different from the rest of the data, may be plotted separately from the main box plot.
Box plots are widely used in statistics and data analysis to visualize the distribution of data. They are particularly useful for comparing multiple data sets or groups of data.
In Figure 16 we generate a box-and-whisker plot, showing the distribution of \texttt{Age} for each of the \texttt{Obesity} levels.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Diabetes,
y = Age
)
) +
geom_boxplot() +
labs(
title = "Age distribution of participants",
subtitle = "Grouped by diabetes level"
) +
ggthemes::theme_economist()
A violin plot is a graphical representation of a data set that combines elements of a box plot and a kernel density plot. It is used to display the distribution of the data, including the median, quartiles, and range, as well as the underlying density of the data.
A violin plot consists of a narrow rectangle with a thicker kernel density plot plotted on the center. The kernel density plot shows the probability density of the data, which is a smoothed version of the histogram of the data. The rectangle represents the interquartile range (IQR) of the data and the median of the data is represented by a line drawn within the rectangle. The lower and upper quartiles are represented by the lower and upper edges of the rectangle, respectively. The minimum and maximum values of the data are represented by the lower and upper ends of the kernel density plot, respectively.
Violin plots are useful for comparing multiple data sets or groups of data and for visualizing the distribution of data. In Figure 17 we show the distribution of \texttt{Age} across the levels of \texttt{Diabetes}, using the geom_violin geometry.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Diabetes,
y = Age
)
) +
geom_violin(
draw_quantiles = c(0.25, 0.5, 0.75)
) +
labs(
title = "Age distribution of participants",
subtitle = "Grouped by diabetes level"
) +
ggthemes::theme_economist()
Scatter plots
A scatter plot is a graphical representation of a data set that displays the values of two variables for each data point as a pair of coordinates on a graph. It is used to visualize the relationship between the two variables and to identify patterns and trends in the data.
To create a scatter plot, you plot each data point as a pair of coordinates on a graph, with the value of one variable on the x-axis and the value of the other variable on the y-axis. The resulting graph will show you the relationship between the two variables, with the slope and direction of the plotted points indicating the strength and type of the relationship.
Scatter plots are particularly useful for visualizing the relationship between two continuous variables, such as measurements or observations. They are also useful for identifying outliers in the data, as points that fall outside the general pattern of the plotted points will be easily visible. Scatter plots are widely used in statistics and data analysis to visualize relationships between variables and to identify patterns and trends in the data.
In Figure 18 we plot the \texttt{Age} and \texttt{CRP} pairs for each observation using the geom_point geometry.
dfr %>% ggplot2::ggplot(
aes(
x = Age,
y = CRP
)
) +
geom_point(
shape = 1,
size = 3,
) +
labs(
title = "Age vs. CRP level",
subtitle = "Includes all participants"
) +
ggthemes::theme_economist()
A scatter plot can be grouped by the levels of a categorical variable. In Figure 19 the \texttt{Age} and \texttt{CRP} variables are grouped by the levels of the \texttt{Obesity} variable using the group argument. An aes argument is now also added to the geom_point function. We specify that different colors and marker shapes must be used for each level of the \texttt{Obesity} variable, by setting the color and the shape arguments.
dfr %>% ggplot2::ggplot(
aes(
x = Age,
y = CRP
)
) +
geom_point(
aes(
color = Obesity,
shape = Obesity
),
size = 3,
) +
labs(
title = "Age vs. CRP level",
subtitle = "Grouped by obesity level"
) +
ggthemes::theme_economist()
In Figure 20 we add a third continuous variable, \texttt{HR}. This is done by setting the size argument. The sizes of the markers are determined by the values of \texttt{HR}.
dfr %>% ggplot2::ggplot(
aes(
x = Age,
y = CRP
)
) +
geom_point(
aes(
color = Obesity,
size = HR
),
shape = 16,
alpha = 0.5
) +
scale_color_manual(values=c('#E69F00', '#56B4E9')) +
labs(
title = "Age vs. CRP level grouped by obesity level",
subtitle = "Marker size indicates heart rate"
) +
ggthemes::theme_clean()
The ggscatterstats function from the ggstatsplot library creates a publication ready plot that adds a histogram in the margin for each of the two continuous variables. The subtitle in Figure 21 also includes results for frequentest inference.
Plotting for categorical data
For categorical data we visualize the frequency of the sample space elements of a variable. The most commonly used plots are bar plots. They can be set to indicate frequency or relative frequency. Bar plot can also be used to visualize more than one categorical variable.
Pie charts can be used to visualize relative frequency too. Due the the difficulty in perceiving small differences in the proportions of slices of a pie chart, they are not recommended as useful plot types to visually represent data accurately.
Bar plot
In Figure 22 we create a simple bar plot, showing the frequency of the three levels of the \texttt{Obesity} variable. Notice the one major visual difference between a histogram and a bar plot. In the latter, the rectangles as separated from each other. The separation is to indicate that the values of a categorical variable are not a continuum. In a histogram, the bars are contiguous, providing a visual representation of a continuous variable.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
)
) +
ggplot2::geom_bar() +
labs(
title = "Distriburtion of diabetes levels"
) +
xlab("Diabetes variable levels") +
ylab("Frequency") +
ggthemes::theme_economist_white()
The rectangles can be filled differently for each level of the categorical variable. Default colors are chosen in Figure 23.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Diabetes
)
) +
ggplot2::geom_bar() +
labs(
title = "Distriburtion of diabetes levels"
) +
xlab("Diabetes variable levels") +
ylab("Frequency") +
ggthemes::theme_economist_white()
We can manually select colors using the scale_fill_manual function. The values argument takes a vector of hexademical values. The plot is shown in Figure 24.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Diabetes
)
) +
ggplot2::geom_bar() +
ggplot2::scale_fill_manual(
values = c(
"#999999", "#999999", "#EEAAAA"
)
) +
labs(
title = "Distriburtion of diabetes levels"
) +
xlab("Diabetes variable levels") +
ylab("Frequency") +
ggthemes::theme_economist_white()
The stat argument for the geom_box function is set the bin by default. It can also be set to idenity. This sums a numerical variable, for which we have to provide the values. In Figure 25 we provide a value for the y argument in the aes statement.
dfr %>% ggplot2::ggplot(
aes(
x = Obesity,
y = Age,
fill = Obesity
)
) +
ggplot2::geom_bar(
stat = "identity"
) +
labs(
title = "Sum of ages for each level of obesity"
) +
xlab("Obesity variable levels") +
ylab("Frequency") +
ggthemes::theme_economist_white()
A second categorical variable can be added to the data visualization. In Figure 26 we visualize both the \texttt{Diabetes} and the \texttt{Obesity} categorical variables. The position argument is set to dodge. This creates bars that are next to each other. When the number of levels are small, we can distinguish the frequencies relative to each other with ease.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Obesity
)
) +
ggplot2::geom_bar(
position = "dodge"
) +
labs(
title = "Frequency of obesity levels per diabetes levels"
) +
ylab("Frequency") +
ggthemes::theme_excel_new()
The bars can also be stacked, shown in Figure 27, where we set the position argument to "stack". Stacking makes it difficult to interpret the absolute frequency of the the bars that are not at the bottom of the stacks. Stacking introduces the same problems that we have with pie charts.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Obesity
)
) +
ggplot2::geom_bar(
position = "stack"
) +
labs(
title = "Frequency of obesity levels per diabetes levels"
) +
ylab("Frequency") +
ggthemes::theme_excel_new()
In Figure 27 it is still possible to consider the top of the stacked bars as visual representation of the frequency of the levels of the variable on the horizontal axis. In Figure 28 we set the position argument to "fill". The bars fill all of the vertical space and we can only consider proportions.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Obesity
)
) +
ggplot2::geom_bar(
position = "fill"
) +
labs(
title = "Proportion of obesity levels per diabetes levels"
) +
ylab("Frequency") +
ggthemes::theme_excel_new()
There is no need to have the joint frequencies on the same plot. The facet_warp function creates separate plots for the levels of the specified variable. In Figure 29 the vars argument is set to Obesity so that we have separate plots for each of the levels of the \texttt{Obesity} levels. The short-hand syntax ~Obesity can also be used. The ncol argument indicates the number of columns, each which will contain a plot. The value is set to the number of unique elements of the \texttt{Obesity} variable. The strip-position argument is set to "left". This prints the specific levels of the \texttt{Obesity} variable on the left of each plot instead of at the top.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Diabetes
)
) +
ggplot2::geom_bar() +
ggplot2::facet_wrap(
vars(Obesity),
ncol = 2,
strip.position = "left"
) +
labs(
title = "Frequency of obesity levels per diabetes levels"
) +
ylab(NULL) + # Set to NULL to allow for variable levels
ggthemes::theme_excel_new()
In Figure 30 the nrow argument is used to produce two plots in vertical orientation. The labeller argument overwrites the values of the levels.
dfr %>% ggplot2::ggplot(
aes(
x = Diabetes,
fill = Diabetes
)
) +
ggplot2::geom_bar() +
ggplot2::facet_wrap(
vars(Obesity),
nrow = 2,
strip.position = "left",
labeller = as_labeller(c(No = "Frequency for normal weight", Yes = "Frequency for obesity"))
) +
labs(
title = "Frequency of obesity levels per diabetes levels"
) +
ylab(NULL) +
xlab("Diabetes") +
ggthemes::theme_excel_new()
dfr %>% ggplot2::ggplot(
aes(
x = Obesity,
fill = Obesity
)
) +
ggplot2::geom_bar() +
ggplot2::facet_wrap(
vars(Grade),
nrow = 3,
strip.position = "top",
labeller = as_labeller(c(
I = "Frequency for grade I",
II = "Frequency for grade II",
III = "Frequency for grade III"
))
) +
labs(
title = "Obesity levels per grade of disease levels"
) +
ylab(NULL) +
ggthemes::theme_excel_new()
Conclusion
The ggplot2 library is perhaps to most-used plotting library in statistics and data science. The types of plots and how to customize them are near endless.
Many other plotting libraries exist. You are urged to read up on some of them.
Lab assignmnemt
[40 points]
Section 1
[10 points]
Select one numerical and one categorical variable in the data set imported in this notebook and generate a box-and-whisker plot that has not been created as example in this notebook.
Section 2
[10 points]
Select two numerical variables and a categorical variable and create a scatter plot (grouped by the levels of the categorical variable) that has not been created as example in this notebook.
Section 3
[20 points]
Look up what a mosaic plot is. Import an appropriate library and create a mosaic plot of the variables \texttt{Ventilation} and \texttt{Obesity}.